Real-time monitoring methods, systems and storage media for soil amendment effects

Abstract

This application relates to the field of soil remediation technology, and discloses a method, system, and storage medium for real-time monitoring of soil remediation effects. The method includes: collecting multi-depth data samples of soil in the vertical direction to construct a stratified soil characteristic dataset; identifying locations of highly heterogeneous sections based on the dataset; injecting a tracer at these locations and imaging in real time to obtain a sequence of infiltration path images; using a random forest model to predict diffusion velocity and blockage point locations, dynamically adjusting injection parameters to eliminate blockage, and obtaining infiltration path data; tracking and quantifying the effects of the soil amendment on each soil layer through time-series analysis to obtain quantitative indicators of the effect; comparing these indicators with the initial dataset to optimize the stratification model and construct a stratified monitoring framework; integrating multi-depth data samples to validate the framework and generating an evaluation report on the soil remediation effect. This invention achieves accurate characterization of soil vertical stratification characteristics and scientific, quantitative evaluation of remediation effects, improving the efficiency of soil amendment use.

Description

Technical Field

[0001] This application relates to the field of soil improvement technology, and in particular to a method, system and storage medium for real-time monitoring of soil improvement effects. Background Technology

[0002] Soil remediation is a core component of ensuring agricultural production and ecological environmental security. With the continuous increase in land use intensity, soil quality degradation and structural heterogeneity are becoming increasingly prominent problems, making soil improvement a key measure to enhance land productivity and achieve sustainable development.

[0003] However, current soil improvement practices face severe challenges, mainly in the following aspects: First, soil exhibits significant differences in physical properties (such as porosity and density) and chemical composition (such as pH and organic matter) along the vertical direction, forming a complex vertical heterogeneous structure. Current technologies lack sufficient understanding of this stratification characteristic, making it difficult to accurately grasp the distribution patterns of characteristics in each soil layer. Second, due to the lack of accurate analysis of soil vertical heterogeneity, soil amendments are difficult to distribute uniformly during application, often resulting in flow hindrance in dense soil layers or heterogeneous interfaces, leading to poor improvement effects in some areas and even resource waste. This uneven penetration not only increases remediation costs but also makes the evaluation of improvement effects lack a scientific basis. Furthermore, existing monitoring methods are mostly limited to surface soil or use overall averaging assessments, failing to quantify and track the actual action process and duration of amendments in soil layers at different depths, resulting in ambiguous effect evaluations and difficulty in guiding the optimization of subsequent improvement strategies.

[0004] Therefore, how to accurately identify the vertical stratification characteristics of soil and, based on this, dynamically optimize the soil amendment penetration process and quantitatively monitor its effects has become a key technical problem that urgently needs to be solved in the field of soil remediation. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide a method, system and storage medium for real-time monitoring of soil improvement effects, which can realize accurate analysis of soil vertical stratification characteristics, dynamic optimization of soil amendment penetration process and quantitative evaluation of its effects, thereby improving the accuracy and stability of soil improvement effects.

[0006] Firstly, this application provides a method for real-time monitoring of soil amendment effects, the method comprising:

[0007] Step 1: Collect multi-depth data samples of soil in the vertical direction using a sensor array to construct a layered soil characteristic dataset;

[0008] Step 2: Identify the locations of highly heterogeneous zones in the soil based on the physical and chemical properties of the stratified soil characteristics dataset;

[0009] Step 3: Inject tracers into the high heterogeneous section to simulate the flow of modifiers and capture their distribution images in real time to obtain a permeation path image sequence;

[0010] Step 4: Based on the permeation path image sequence, predict the diffusion rate and stagnation point location of the tracer, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier and obtain updated permeation path data.

[0011] Step 5: Based on the updated infiltration path data, track and quantify the changes in the effect of the amendment in each soil layer, and obtain quantitative indicators of the effect.

[0012] Step 6: Compare the quantitative indicators of the effect with the stratified soil characteristics dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring accordingly.

[0013] Step 7: Integrate multi-depth data samples to validate the hierarchical monitoring framework and generate the final vertical hierarchical effect evaluation report.

[0014] Secondly, this application provides a real-time monitoring system for soil amendment effects, the system comprising:

[0015] The data acquisition module is used to collect multi-depth data samples of soil in the vertical direction through a sensor array to construct a layered soil characteristic dataset;

[0016] The segment identification module is used to identify the location of highly heterogeneous segments in the soil based on the physical and chemical properties of the stratified soil characteristic dataset.

[0017] The image acquisition module is used to inject tracers into highly heterogeneous sections to simulate the flow of modifiers and capture their distribution images in real time to obtain a sequence of permeation path images.

[0018] The path optimization module is used to predict the diffusion rate and hindrance point location of the tracer based on the permeation path image sequence, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier and obtain updated permeation path data.

[0019] The effect quantification module is used to track and quantify the changes in the effect of the amendment in each soil layer based on the updated infiltration path data, and obtain quantitative indicators of the effect.

[0020] The model update module is used to compare the quantitative indicators of the effect with the stratified soil characteristics dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring accordingly.

[0021] The validation report module is used to integrate multi-depth data samples to validate the hierarchical monitoring framework and generate a final vertical hierarchical effect evaluation report.

[0022] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for real-time monitoring of soil improvement effects.

[0023] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows:

[0024] 1. Through multi-depth sensor arrays and layered data analysis, the physicochemical properties of soil in the vertical direction are accurately characterized, and the location of highly heterogeneous sections can be accurately identified. This overcomes the shortcomings of traditional methods in understanding soil stratification characteristics and accurately characterizes soil vertical heterogeneity.

[0025] 2. Based on machine learning models for predicting permeation paths and dynamically processing hindrance points, injection parameters can be dynamically adjusted during the modifier injection process. This enables proactive intervention and optimization of the modifier flow process, significantly improving the uniformity of modifier distribution in the vertical direction, optimizing modifier permeation uniformity, and effectively avoiding resource waste.

[0026] 3. Through time-series analysis and the construction of quantitative indicators, a complete evaluation system from infiltration path to effect was established, which can accurately quantify the intensity and duration of the effect of the soil conditioner in each soil layer, providing a scientific basis for effect evaluation and realizing quantitative evaluation of effect.

[0027] 4. Through iterative optimization of the hierarchical model and construction of the monitoring framework, an intelligent monitoring system with self-learning capabilities has been formed. It is not only suitable for evaluating the effect of a single improvement, but also provides decision support for continuous optimization of subsequent soil remediation, thereby improving the intelligence level of the monitoring system. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of the method for real-time monitoring of soil amendment effects according to this application;

[0030] Figure 2 This is a schematic diagram of the structure of the real-time monitoring system for soil improvement effects proposed in this application. Detailed Implementation

[0031] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of the real-time monitoring method for soil improvement effects provided by the present invention. The flowchart specifically includes the following steps:

[0033] Step 1: Collect multi-depth data samples of soil in the vertical direction using a sensor array to construct a layered soil characteristic dataset.

[0034] In one specific embodiment, the process of performing step 1 may specifically include the following steps:

[0035] A stratified sampling method was used to divide the vertical soil profile into topsoil, subsoil and subsoil, and data samples of each soil layer were collected by a sensor array.

[0036] The data samples were preprocessed, and then the physical and chemical parameters of each soil layer were extracted.

[0037] An initial stratified soil property dataset is constructed using physical and chemical parameters. This dataset contains information on the vertical distribution of properties of each soil layer.

[0038] The initial stratified soil property dataset was standardized to obtain structured standard data;

[0039] Structured standard data is stored as a multidimensional matrix, where rows represent different soil layers and columns represent different parameters;

[0040] Feature extraction is performed on the multidimensional matrix to determine the correlation between the properties of each soil layer;

[0041] The initial stratified soil property dataset is preliminarily stratified and validated based on the correlation. If the validation is successful, the structured standard data will be used as the stratified soil property dataset.

[0042] Specifically, the soil profile is structurally divided into topsoil, subsoil, and subsoil layers according to soil science standards. A sensor array is pre-positioned within the profile, containing multiple probes, each responsible for monitoring a specific depth range. For example, in an agricultural soil scenario, the topsoil is defined as 0–20 cm, the subsoil as 20–50 cm, and the subsoil as below 50 cm. Once activated, the probes collect raw data samples from each soil layer, including information on soil moisture, density, porosity, pH, and organic matter content.

[0043] The collected raw data samples then enter the preprocessing stage. This stage uses filtering algorithms to remove invalid data points introduced by environmental interference or equipment noise, and interpolation methods are used to fill in missing data to ensure the integrity of the data samples. The preprocessed data is then input into the feature extraction module, which extracts physical and chemical parameters representing the soil condition from the samples. Physical parameters include at least moisture, porosity, and density, while chemical parameters cover nitrogen, phosphorus, and potassium content, pH value, and organic matter content. Preferably, the extraction process requires setting corresponding extraction thresholds according to the parameter type; for example, the porosity extraction range is set to 0.2 to 0.6, and the pH value extraction range is set to 4.5 to 8.5 to avoid invalid data entering subsequent processes.

[0044] The extracted physical and chemical parameters were organized according to soil layer categories to construct an initial stratified soil characteristic dataset. The dataset is indexed by depth level, recording the vertical distribution of characteristics for each soil layer. For example, the surface layer has a porosity of 0.4, a density of 1.2 g / cm³, a pH of 6.5, and an organic matter content of 2.0%; the subsoil layer has a porosity of 0.3, a density of 1.4 g / cm³, a pH of 6.0, and an organic matter content of 1.5%; and the subsoil layer has a porosity of 0.2 and a density of 1.6 g / cm³. 3 With a pH value of 5.5 and an organic matter content of 1.0%, this organization method allows for a direct presentation of the differences in characteristics among soil layers, recording the characteristic distribution information of each soil layer in the vertical direction.

[0045] Because different parameters, such as porosity and pH, have different dimensions and numerical ranges, direct comparison or analysis can introduce bias. Therefore, it is necessary to standardize the initial stratified soil property dataset. Preferably, the standardization process uses a min-max scaling method. After this step, the original, dimensionally inconsistent initial dataset is transformed into structured standard data. This standard data is stored in the form of a multidimensional matrix, where each row uniquely corresponds to a soil layer (e.g., the first row represents the topsoil, the second the subsoil, and the third the subsoil), and each column corresponds to a specific soil parameter (e.g., the first column is porosity, the second is pH, and the third is organic matter content). This row-column structure allows for the precise encoding and location of soil vertical stratification and attribute information.

[0046] To further explore the intrinsic relationships among soil layer properties, feature extraction was performed on the constructed multidimensional matrix. Principal component analysis (PCA) was employed, which identifies the main directions of data variation by calculating the eigenvalues ​​and eigenvectors of the data covariance matrix, thus transforming multiple potentially correlated parameters into a few linearly uncorrelated principal components. After extracting the principal components, Pearson correlation coefficients were calculated among these principal component features to determine correlations such as porosity and pH. By analyzing the correlation coefficient matrix, the statistical correlations between various soil layer properties can be determined; for example, a strong positive correlation between porosity and pH may be found in specific soil layers.

[0047] Preliminary stratification validation is performed on the initial stratified soil property dataset based on correlation. A correlation difference threshold (e.g., 0.5) is set, and the correlation coefficient differences of various parameters between adjacent soil layers are compared. If the correlation or difference indices of all parameters between all adjacent soil layers do not exceed the corresponding threshold, it indicates that the stratification result is reasonable and can capture vertical abrupt changes in soil properties, thus passing the validation. At this point, the previously generated structured standard data is officially confirmed as a stratified soil property dataset that can be used for subsequent analysis. If the validation fails, a backup plan is initiated, such as re-dividing the soil layers using fixed depth intervals and updating the dataset to ensure data availability.

[0048] This data processing workflow addresses the technical challenges of insufficient understanding of soil vertical heterogeneity and the difficulty in accurately grasping the distribution patterns of characteristics across soil layers. Through a series of operations from raw sampling to the construction of a standardized dataset, coarse field measurement data is transformed into clean, well-organized, dimensionally consistent, and preliminarily validated structured information. This data transformation and enhancement lays a reliable data foundation for accurately identifying highly heterogeneous zones, enabling the application of soil amendments based on a deep understanding of the soil's vertical structure. This, in turn, helps to address the resource waste and poor effectiveness caused by uneven infiltration at the source.

[0049] Step 2: Identify the locations of highly heterogeneous zones in the soil based on the physical and chemical properties of the stratified soil characteristics dataset.

[0050] In one specific embodiment, the process of performing step 2 may specifically include the following steps:

[0051] The stratified soil property dataset is input into a pre-trained support vector machine model. The support vector machine model is used to classify the physical properties and chemical composition in the dataset to obtain the classification results of each soil layer.

[0052] Based on the classification results, calculate the characteristic differences in physical and chemical properties between adjacent soil layers;

[0053] Determine whether the characteristic difference value exceeds the preset difference threshold. If so, mark the corresponding soil segment as a high heterogeneous segment.

[0054] For the marked highly heterogeneous segments, extract their vertical position information;

[0055] A distribution map of highly heterogeneous segments is generated based on the location information. The distribution map is then used to accurately locate the highly heterogeneous segments and obtain their positions.

[0056] Specifically, a stratified soil property dataset is input into a pre-trained Support Vector Machine (SVM) model. This model is a supervised learning algorithm trained on historical soil data, which includes annotations of different soil layers and their physical and chemical properties. The SVM model works by finding a hyperplane in a high-dimensional feature space that maximizes the margin between soil samples of different categories. During classification, the model receives input data, such as a data vector containing porosity, density, pH, and organic matter content, and maps it to a high-dimensional space. Using kernel functions, such as radial basis functions, the nonlinear similarity between samples is calculated to determine the soil layer category to which the data vector belongs. The output is a classification label for the topsoil, subsoil, or subsoil layer.

[0057] Preferably, during the model training phase, internalizing soil types (such as sandy soil, clay soil, and loam) can improve the "background knowledge" or "intrinsic model parameters" for vertical stratification accuracy. For example, when analyzing a sandy soil profile, the model will use its learned "sand characteristic pattern" to classify low-density, high-porosity areas as the surface layer and areas with slightly increased density and slightly decreased porosity as the subsoil layer; when analyzing a clay soil profile, the model will switch to the "clay characteristic pattern," classifying relatively loose areas as the surface layer and very dense areas as the subsoil layer or subsoil layer.

[0058] After obtaining the classification results for each soil layer, the characteristic difference value between adjacent soil layers is calculated. This calculation is based on the physical and chemical property information contained in the classification results and can be performed using the Euclidean distance formula. For a soil layer containing N characteristics, the characteristic difference value Dab between two adjacent soil layers A and B can be calculated using the formula... We obtain p, where p an and p bn These represent the standardized values ​​of soil layer A and soil layer B on the nth property, respectively. This calculation quantifies the combined deviation of adjacent soil layers across multiple properties.

[0059] The calculated characteristic difference value is compared with a preset difference threshold. This threshold is an empirical value derived from the analysis of a large amount of soil profile data and is used to define whether the soil properties have changed significantly. For example, in the agricultural soil improvement scenario, considering the different responses of amendments to soils with different characteristics, the threshold is set to 0.4. The judgment logic is: if the characteristic difference value is greater than the preset difference threshold, it indicates that there is a property abrupt interface between these two adjacent soil layers. Soil sections that meet this condition will be marked as highly heterogeneous sections. For example, if the characteristic difference value between the subsoil and the subsoil is 0.55, exceeding the threshold of 0.4, then the boundary between the subsoil and the subsoil and the area above the subsoil are marked as highly heterogeneous sections; while the characteristic difference value between the surface layer and the subsoil is 0.25, not exceeding the threshold, and this area is not marked.

[0060] For each marked highly heterogeneous segment, vertical location information is extracted from the depth information associated with the stratified soil property dataset. The extraction is based on the depth labels corresponding to each parameter in the dataset. For example, if the marked highly heterogeneous segment corresponds to a depth range of 45 to 60 cm, the starting depth of 45 cm and the ending depth of 60 cm in the vertical direction of the segment are recorded. Using the extracted location information of all highly heterogeneous segments, combined with the soil vertical profile coordinate system, the starting depth, ending depth, and corresponding soil layer category information are imported into a plotting tool to generate a distribution map of the highly heterogeneous segments. The distribution map uses depth as the vertical axis and the soil horizontal profile as the horizontal axis, filling the coordinate area corresponding to the highly heterogeneous segment with a specific color (such as red), and marking the average values ​​of the main physical and chemical parameters within the segment (such as porosity 0.25, pH value 5.6). The distribution map allows for a visual view of the specific location of highly heterogeneous sections in the vertical profile, enabling further precise positioning of these sections. For example, it can be determined that the center of the highly heterogeneous section is at a depth of 52 cm, and its horizontal distribution range is 10 cm on each side of the central axis of the soil profile.

[0061] This data processing workflow addresses the technical challenges of uneven soil amendment distribution and poor performance in certain areas due to insufficient understanding of soil vertical heterogeneity. By using a support vector machine model to intelligently classify soil layers and quantify the characteristic differences between adjacent soil layers, it can automatically and objectively identify critical zones where soil properties undergo abrupt changes. This data-driven identification method overcomes the subjectivity and uncertainty of traditional experience-based judgments. Clearly identifying the location of highly heterogeneous zones provides target areas for subsequent targeted injection of tracers and simulation of amendment flow, avoiding tracer injection deviations caused by location ambiguity. This allows resources to be concentrated on the soil sections most in need of attention and treatment, thus strategically optimizing the amendment penetration path and mitigating the problems of uneven penetration and resource waste caused by indiscriminate application.

[0062] Step 3: Inject tracers into the high heterogeneous section to simulate the flow of modifiers and capture their distribution images in real time to obtain a permeation path image sequence.

[0063] In one specific embodiment, the process of performing step 3 may specifically include the following steps:

[0064] The tracer injection point is determined based on the location of the highly heterogeneous section;

[0065] The flow of soil amendments in a vertical soil profile is simulated by injecting tracers at injection points using an injection device.

[0066] The distribution of the tracer in the vertical path is dynamically captured using a real-time imaging device, and multiple distribution images are obtained based on the dynamically captured data.

[0067] Arrange the distribution images in a temporal sequence to construct an initial infiltration path image sequence;

[0068] The permeation path image sequence is obtained by denoising the initial permeation path image sequence.

[0069] Specifically, through physical simulation and imaging technology, the invisible flow process of soil amendments is transformed into a visualized and analyzable data sequence.

[0070] The tracer injection points are set at locations in the soil corresponding to the vertical projection of the highly heterogeneous section. This aims to allow the tracer to flow vertically downwards through the target section under the influence of gravity and external pressure, effectively simulating the infiltration environment that the amendment might encounter in practical applications. The number of injection points can be one or more, depending on the horizontal distribution range of the highly heterogeneous section, to ensure that the tracer cloud effectively sweeps across the entire target area. For example, if the highly heterogeneous section is defined as a region with a vertical depth of 45 to 60 cm and a horizontal range of 10 cm on each side of the soil profile's central axis, then at the surface layer (0 cm depth), the main injection point can be set at the center of the vertical projection of this region. Simultaneously, an auxiliary injection point can be added on each side of this center point, for example, at a horizontal interval of 10 cm, forming a multi-point injection layout at the surface layer. This improves the uniformity and representativeness of the tracer distribution within the deeper target section.

[0071] The injection process is performed by specialized injection equipment, such as a high-pressure injection pump that precisely controls flow and pressure. This equipment injects a visible or detectable tracer, such as a fluorescent dye or a radioactive isotope solution, into the soil profile at a constant rate from a predetermined injection point. This step directly simulates the application process of liquid soil conditioners. The tracer begins its complex transport within the soil pores, its path and velocity profoundly influenced by soil pore structure, water saturation, and, particularly, the physical properties of highly heterogeneous sections.

[0072] To capture this dynamic process, real-time imaging equipment was deployed to continuously monitor the distribution of the tracer along its vertical path. This imaging equipment, such as an X-ray computed tomography (CT) system or ground-penetrating radar, was positioned to one side or around a soil profile and scanned the target profile at preset time intervals, such as once per second. Each scan generated a two-dimensional or three-dimensional distribution image, reflecting the differences in tracer concentration or density at different points within the soil profile. This dynamic capture continued for a period of time, such as ten minutes, to ensure coverage of the complete cycle from the start of tracer infiltration to the point where its flow stabilized or changed significantly. Through this continuous scanning, a series of raw distribution images, arranged chronologically, recording the spatial distribution of the tracer were obtained.

[0073] These temporally consecutive raw distribution images constitute the initial infiltration path image sequence. However, these raw images typically contain various types of noise, such as scattering caused by heterogeneous particles in the soil matrix, electronic noise inherent in the imaging equipment, or environmental electromagnetic interference. This noise can obscure the actual flow trajectory of the tracer, interfering with the accuracy of subsequent analysis. Therefore, the initial infiltration path image sequence needs to undergo a denoising process using digital image processing algorithms, such as median filtering. Median filtering is a non-linear technique that calculates the median of the gray values ​​of all pixels in the neighborhood of each pixel in the image and replaces the original pixel value with this median. This process effectively filters out impulse noise and speckle noise while preserving image edge information relatively well. Subsequently, to further highlight the flow front of the tracer, edge enhancement algorithms may also be applied. After this series of denoising and enhancement processes, the tracer flow trajectory, which was previously obscured by noise in the initial sequence, becomes clear and coherent. The data at this point is defined as a high-quality infiltration path image sequence, which clearly reveals the infiltration path of the tracer in the vertical soil profile over time.

[0074] This data processing workflow solves the technical problems of uneven penetration and ambiguous effect evaluation caused by the inability to grasp the actual vertical distribution of the modifier. By injecting targeted tracers and performing real-time imaging above highly heterogeneous sections, the flow process of the modifier is transformed from speculation into visualized image data. Denoising processing improves the quality of the image data, ensuring the reliability of the basis for subsequent analysis. The obtained permeation path image sequence provides direct and dynamic visual evidence for the next step of predicting diffusion velocity and identifying hindrance points. This allows the system to diagnose problems based on real flow conditions rather than theoretical models, thus enabling dynamic adjustment of injection parameters and optimization of the actual distribution of the modifier. This directly addresses the resource waste and poor results caused by unclear permeation paths and unclear hindrance points.

[0075] Step 4: Based on the permeation path image sequence, predict the diffusion rate and stagnation point location of the tracer, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier, and obtain updated permeation path data.

[0076] In one specific embodiment, the process of performing step 4 may specifically include the following steps:

[0077] The permeation path image sequence is input into a pre-trained random forest model, and the random forest model extracts features from the permeation path image sequence to predict the diffusion rate of the tracer.

[0078] The flow resistance of the tracer in each soil layer is determined based on the diffusion rate;

[0079] Determine whether the flow resistance exceeds the preset resistance threshold; if so, determine that there is a bottleneck.

[0080] For the identified bottleneck points, their coordinates are extracted from the permeation path image sequence to obtain the location of the bottleneck points;

[0081] If the location information of the hindrance point shows that the hindrance point is located in the bottom soil layer, then adjust the injection parameters, including the injection pressure and the injection rate.

[0082] The tracer was re-injected according to the adjusted injection parameters, and a new permeation path image sequence was captured.

[0083] The new permeation path image sequence is analyzed to determine whether the blockage point has been eliminated. If so, the new permeation path image sequence is used as the updated permeation path data. If not, the injection parameters are iteratively adjusted until the blockage point is eliminated.

[0084] Specifically, the infiltration path image sequence is input into a pre-trained random forest model, an ensemble learning algorithm composed of numerous decision trees. During model training, it is trained using historically accumulated soil infiltration image sequences under different soil types and injection parameters, along with corresponding measured flow velocity data, to learn the mapping relationship from image features to physical velocity. When a new image sequence is input, the model extracts features, including but not limited to the pixel displacement of the tracer front between consecutive frames in the sequence, the rate of change of front morphology, and spatiotemporal statistics of image texture. Multiple decision trees within the model independently analyze and vote on these features, ultimately outputting a predicted value for the average diffusion velocity of the current tracer in the soil vertical profile, for example, a predicted velocity of 0.3 mm per second.

[0085] Based on the predicted diffusion rate, the flow resistance of the tracer in each soil layer is further calculated. Flow resistance can be modeled as a function of the diffusion rate. For example, the flow resistance calculation is based on the derivation of Darcy's law. Where R is the flow resistance, k is the soil permeability coefficient (e.g., 5×10^-5 m / s for the surface layer, 2×10^-5 m / s for the subsoil, and 8×10^-6 m / s for the base layer, obtained through actual measurements based on soil type), L is the soil layer thickness, and v is the predicted diffusion velocity (unit converted to m / s). This calculation transforms the kinematic information (velocity) observed in the image into a physical parameter (resistance) reflecting the degree to which the soil medium impedes the passage of fluid.

[0086] The calculated flow resistance R is then compared with a preset resistance threshold Rth. This threshold is an empirical value determined based on a large amount of soil permeability data and is used to identify the boundary between normal permeability and severe obstruction. The judgment logic is as follows: if the flow resistance R in a certain soil area is greater than the preset threshold Rth, then an obstruction point is determined to exist in that area.

[0087] For the identified blockage points, their spatial coordinates are extracted from the image sequence of the infiltration path that led to the determination. Through image registration and coordinate mapping techniques, the pixel location of the blockage point in the image is converted into its three-dimensional spatial coordinates in the actual soil profile, thereby obtaining the specific location information of the blockage point. For example, in the subsoil layer region (depth 50–80 cm), in images from frame 40 (20 minutes) to frame 60 (30 minutes), the tracer diffusion boundary remains at a depth of 65 cm, without moving to deeper layers, and the tracer concentration at this location is concentrated (grayscale value is 20% higher than the surrounding area). Therefore, the coordinates of the blockage point are determined to be at a vertical depth of 65 cm and a horizontal center position (consistent with the horizontal coordinates of the injection point). The coordinate extraction process combines the temporal changes and concentration distribution characteristics of the images to ensure the accurate location of the blockage point.

[0088] After acquiring location information, a decision is made based on a preset strategy. If the location information of the hindrance point indicates that it is located in the subsoil layer, the parameter adjustment procedure is initiated. The adjustment of injection parameters follows specific control logic. For example, to overcome the typically high density and cohesion of the subsoil layer, the system instructs the injection device to increase the injection pressure while appropriately reducing the injection rate. Increasing the pressure enhances the tracer's penetration into dense soil, while reducing the rate prevents excessive accumulation of the tracer in the hindrance point area, balancing infiltration efficiency and diffusion uniformity. The specific adjustment amount can be obtained based on a simple lookup table method or proportional control.

[0089] Based on the adjusted injection parameters, the injection device is controlled to re-inject the tracer into the soil, and a real-time imaging device is simultaneously activated to capture new permeation path image sequences generated by the tracer flow under the new parameters. Next, this new set of permeation path image sequences is analyzed, with the core objective of determining whether the previously identified bottlenecks have been eliminated. The analysis process includes calculating whether the flow resistance in the region near the original bottleneck coordinates has decreased below the threshold Rth, or determining through image analysis whether the tracer front has continuously passed through the region. If it is determined that the bottleneck has been eliminated, the new permeation path image sequence acquired in this round is marked as the updated permeation path data, representing the ideal permeation state of the optimized amendment. If it is determined that the bottleneck still exists, the system automatically initiates a new round of iterative adjustments: based on the analysis of the current new sequence, the injection parameters are fine-tuned again (e.g., further increasing pressure or decreasing rate), and then the injection, capture, and judgment process is executed again, forming a closed loop of "adjustment-injection-verification" until the bottleneck is eliminated. The permeation path image sequence obtained after successfully eliminating the bottleneck is then determined as the updated permeation path data.

[0090] This data processing and execution workflow directly addresses the technical problem of uneven soil amendment penetration and blockage in specific areas caused by soil vertical heterogeneity. By predicting physical parameters from image sequences using a machine learning model, quantitative diagnosis of the penetration process is achieved. Operating parameters are dynamically adjusted based on the quantitative diagnosis results, and the adjustment effect is verified in real-time through a new round of imaging, forming a closed-loop control system. This dynamic optimization mechanism ensures that the amendment can overcome flow obstacles in the soil and achieve a more uniform vertical distribution, thus directly addressing the challenges of resource waste and poor deep soil amendment effects caused by unclear penetration paths and passive application.

[0091] Step 5: Based on the updated infiltration path data, track and quantify the changes in the effect of the amendment in each soil layer, and obtain quantitative indicators of the effect.

[0092] In one specific embodiment, the process of performing step 5 may specifically include the following steps:

[0093] The updated permeability path data were processed using time-series analysis to extract the distribution characteristics of the amendment in each soil layer.

[0094] Based on the distribution characteristics, calculate the duration of action of the soil conditioner in each soil layer;

[0095] By weighted analysis of the duration of action, the intensity of action of the amendment in each soil layer was determined;

[0096] For each soil layer, obtain the corresponding quantitative value of the effect based on its corresponding action intensity and duration;

[0097] The quantitative values ​​of the effects corresponding to each soil layer are standardized to obtain the final quantitative indicators of the effects of each soil layer.

[0098] Specifically, the updated infiltration path data is a sequence of denoised tracer distribution images. The time series analysis method uses an autoregressive integral moving average (ARIMA) model. First, the stationarity of the tracer concentration data for each soil layer in each frame of the image is tested. The ADF test (with a significance level set to 0.05) is used to determine whether the data is stationary. If it is not stationary, first-order differencing is performed to make the data meet the stationarity requirements. Then, an ARIMA (p, d, q) model is constructed for the stationary data, where p=2 (autoregressive order), d=1 (difference order), and q=1 (moving average order). The distribution features are extracted by model fitting. The distribution features include at least the peak tracer concentration, the time of occurrence of the peak concentration, the concentration half-life, and the diffusion range for each soil layer.

[0099] The duration of action of the soil amendment in each soil layer is calculated. The duration is defined as the time it takes for the tracer concentration to decrease from exceeding a detection threshold to falling below it. The calculation first extracts the curves showing the tracer concentration change over time for each soil layer from the distribution characteristics. Then, the time when the concentration first exceeds the detection threshold and the time when it first falls below the detection threshold are determined; the difference between these two times is the duration of action. This duration parameter reflects the potential time window for contact and reaction between the amendment and the soil medium. Furthermore, based on the distribution characteristics, a curve showing the average concentration of the amendment in each soil layer as a function of time is reconstructed or fitted; this is the concentration-time curve. This concentration-time curve quantitatively and intuitively characterizes the complete kinetic process of the amendment entering, reaching its peak concentration, remaining, and ultimately decaying or migrating out.

[0100] A weighted analysis was performed on the calculated duration of action for each soil layer to determine the intensity S of the amendment's effect in each layer. The weighted analysis introduced weighting coefficients reflecting soil depth and expected improvement difficulty, which typically decrease with increasing soil depth. For example, considering agricultural soil improvement needs, the importance coefficients for the topsoil (directly affecting crop root growth) were w1 = 0.4, the subsoil (affecting root expansion and nutrient storage) were w2 = 0.35, and the subsoil (affecting deep water and nutrient supply) were w3 = 0.25, satisfying w1 + w2 + w3 = 1. The intensity of action was obtained by multiplying the duration of action by its corresponding soil layer weight. This weighting method assigns a higher intensity value to maintaining the same duration of action in deeper, more difficult-to-manage soils, making intensity a composite indicator combining the time dimension and the dimension of management challenge.

[0101] Preferably, a concentration peak correction factor is introduced when calculating the action intensity. The correction factor is the ratio of the concentration peak of each soil layer to the maximum concentration peak of all soil layers. The action intensity is obtained by multiplying the action duration by the corresponding soil layer weight and the correction factor.

[0102] For each soil layer, based on its corresponding action intensity and duration, an initial effect quantification value Q is obtained through a combination function. This combination function can be a multiplicative model Q=S×T or an additive model. ,in, and This is a coefficient that adjusts the relative importance of the two parameters. The multiplicative model emphasizes the synergistic effect of intensity and time, while the additive model provides a linear combination method. The initial effect quantification values ​​corresponding to each soil layer are standardized to eliminate the influence of dimensions and facilitate cross-soil layer comparisons. After this processing, the final effect quantification indexes of each soil layer are obtained. These indexes are dimensionless values ​​between 0 and 1, intuitively reflecting the relative strength of the amendment's effect in different soil layers.

[0103] This data processing workflow solves the technical problem of being unable to quantify the effects of soil amendments at different soil depths. By extracting physical features from dynamic image data and transforming these features into duration and intensity parameters with clear physical meaning, standardized quantitative indicators are generated, allowing the vertical gradient of the amendment's effect to be objectively measured and compared. This quantitative capability overcomes the ambiguity of traditional assessment methods, providing direct data support for judging the effectiveness of amendment measures and identifying weak links in soil remediation, thereby enabling targeted optimization of resource allocation and improving the overall efficiency of soil improvement.

[0104] Step 6: Compare the quantitative indicators of the effect with the stratified soil characteristic dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring accordingly.

[0105] In one specific embodiment, the process of performing step 6 may specifically include the following steps:

[0106] The quantitative indicators of the effects of each soil layer are aligned layer by layer with the initial characteristic values ​​of the corresponding soil layers in the stratified soil characteristic dataset.

[0107] Calculate the difference between the quantitative indicator of the effect and the initial characteristic value;

[0108] If the difference value is higher than the preset optimization threshold, the soil stratification model is iteratively optimized to obtain the optimized stratification model.

[0109] A layered monitoring framework is constructed based on the optimized layered model. The layered monitoring framework includes monitoring modules corresponding to each soil layer and a real-time data input interface.

[0110] The applicability of the hierarchical monitoring framework was determined by functional verification.

[0111] The validated stratified monitoring framework is associated and stored with the stratified soil property dataset, and a data synchronization and update mechanism is established.

[0112] By mapping data to the hierarchical monitoring framework, a hierarchical monitoring strategy is generated.

[0113] Specifically, the quantitative indicators of the effects of each soil layer are aligned layer by layer with the initial characteristic values ​​(physical and chemical parameters) of the corresponding soil layer in the stratified soil characteristic dataset constructed in step 1. This process is carried out on the same parameter dimension of the same soil layer. For example, the quantitative indicators of the effects of the surface layer are placed in the same comparison system as the initial porosity of the surface soil.

[0114] All initial characteristic values ​​of any soil layer are read to form a feature vector. These initial characteristic values ​​are then input into a prediction function, which is the embodiment or extension of the soil stratification model in the dimension of effect prediction. Through a predefined operational rule, it maps multiple initial characteristic values ​​to a single predicted effect value. The predicted effect value is the "theoretical estimate" of the soil improvement effect based on the initial characteristic values. Its core is to quantify the correlation between initial characteristics and improvement effects through the model's built-in operational rules. The predicted effect value is compared with the quantitative indicator of the effect across that soil layer to obtain a relative deviation value. The relative deviation values ​​corresponding to each layer are weighted and summed to obtain the difference between the quantitative indicator of the effect and the initial characteristic values.

[0115] The calculated difference value is compared with a preset optimization threshold, which defines the sensitive boundary for model optimization. If the difference value is higher than the preset optimization threshold, it indicates that there is a significant deviation between the soil stratification model's prediction of the improved soil state and the actual observation results, and the model's accuracy needs to be improved. Therefore, iterative optimization of the soil stratification model is triggered. This soil stratification model is the initial soil stratification model generated synchronously after successful validation during the construction of the stratified soil characteristic dataset in step 1. The core parameters of this model are derived from the feature analysis results of the dataset. In addition, the model also includes soil layer classification rules, characteristic thresholds, etc., all of which are derived from the distribution patterns of physical and chemical parameters and soil layer correlations in the stratified soil characteristic dataset. The core purpose of model optimization is to ensure that the model always accurately reflects the vertical stratification characteristics of the soil, providing a reliable basis for the subsequent stratification monitoring framework. Specifically, after triggering optimization, the system prepares a batch of training data, where each sample contains a soil characteristic vector (input) and its corresponding validated soil layer label (actual observation value). The gradient descent algorithm starts working: it feeds this batch of input vectors into the current soil stratification model to obtain a batch of predicted soil layer labels (model output). The algorithm then calculates a loss function that quantifies the overall difference between the predicted and validation labels. By calculating the gradient of the loss function with respect to the model's internal parameters, such as the weights and biases related to the decision boundary, the algorithm determines the direction and magnitude of parameter updates. Through multiple iterations, these parameters are gradually adjusted until the model's predicted output infinitely approximates the true validation label, thus optimizing the soil stratification model and ensuring it more accurately reflects the true vertical structure of the soil.

[0116] A layered monitoring framework is constructed based on the optimized layered model. This framework is a software-defined architecture, the core of which is to encapsulate the optimized model logic into independent monitoring modules corresponding to each soil layer (topsoil, subsoil, and subsoil). Each module is responsible for processing the characteristic data and seepage path analysis of its corresponding soil layer. The framework also includes a real-time data input interface, which defines the data format and communication protocol to ensure that new data streams from the sensor array and imaging equipment can be continuously received and distributed to each monitoring module for processing.

[0117] After construction, the stratified monitoring framework undergoes functional verification to determine its applicability. The verification process involves inputting a set of known, representative test data into the framework, such as a sequence of historical infiltration path images, and checking whether the framework can correctly call its internal model to perform stratified analysis and output logically consistent blockage point states and stratification results. If the framework's output matches the expected results within a preset tolerance range, its basic functions are confirmed to be normal and its applicability verified. The verified stratified monitoring framework is then associated and stored with the stratified soil characteristic dataset, and a data synchronization and update mechanism is established between the two. This mechanism ensures that when new monitoring data leads to further optimization of the stratified model, the associated initial characteristic dataset also receives corresponding version tags or metadata updates, maintaining data traceability consistency.

[0118] By mapping data to the hierarchical monitoring framework of associated storage, a hierarchical monitoring strategy is generated. This data mapping associates the parameters of each monitoring module within the framework with the amendment injection strategy. Data mapping is an information transformation process that converts the abstract monitoring modules, model parameters, and interface specifications in the framework into specific, executable instructions and rules. For example, the subsoil monitoring module is mapped to "When a highly heterogeneous section is identified in the subsoil, the injection pressure is recommended to be adjusted to 1.5 times the baseline value"; the data input interface is mapped to "Receive depth sampling data from the sensor array every 30 seconds." Furthermore, the porosity parameter of the surface monitoring module is mapped to the injection pressure adjustment rule (for every 0.1 increase in porosity, the injection pressure decreases by 0.05 MPa), the subsoil pH parameter is mapped to the amendment concentration adjustment rule (for every 0.1 decrease in pH, the amendment concentration increases by 0.02%), and the subsoil density parameter is mapped to the injection rate adjustment rule (for every 0.1 increase in density, the injection rate decreases by 0.2 mL / min). These specific rules and parameters together constitute a stratified monitoring strategy document to guide future soil improvement monitoring operations.

[0119] This data processing and system construction process addresses the technical problems of limited applicability and insufficient long-term reliability of soil improvement monitoring frameworks due to a lack of adaptive optimization capabilities. By feeding back performance evaluations to the model optimization stage, a closed loop is formed, from practice to understanding and back to practical improvement. This iterative mechanism based on data feedback enables the monitoring system to adapt to changes in different soil conditions and improvement stages, enhancing its robustness. The constructed framework with real-time interfaces and its derived specific strategies transform the optimized model capabilities into a sustainable monitoring operation guide, effectively addressing the challenge of traditional static methods gradually losing their effectiveness in complex and dynamic soil environments.

[0120] Step 7: Integrate multi-depth data samples to validate the hierarchical monitoring framework and generate the final vertical hierarchical effect evaluation report.

[0121] In one specific embodiment, the process of performing step 7 may specifically include the following steps:

[0122] Input multi-depth data samples into a validated hierarchical monitoring framework;

[0123] By using a stratified monitoring framework, multi-depth data samples are analyzed in a stratified manner to generate stratified results that include the distribution of soil layer characteristics and the location of highly heterogeneous sections.

[0124] Based on the stratification results, a simulation scenario is constructed. In the simulation scenario, a virtual tracer is injected to reproduce and track the entire process of the modifier's action.

[0125] Based on the reproduction and tracking process, the simulated duration and intensity of the amendment in each soil layer were extracted as data on the effect of the amendment in each soil layer.

[0126] A comprehensive analysis of the effect data is conducted to generate a vertical stratification effect assessment report on the soil improvement effect.

[0127] Specifically, the independent multi-depth data samples collected in step 1, which were not used for framework construction, are input into the previously functionally validated hierarchical monitoring framework. These data samples serve as a test set to evaluate the framework's generalization ability and reliability.

[0128] Upon receiving input data, the framework activates its embedded, optimized soil stratification model to perform automated stratification analysis on multi-depth data samples. This analysis process simulates step 2, calculating and classifying data to output a stratification result containing information on the distribution of soil layer characteristics and the location of highly heterogeneous sections. This stratification result directly reflects the framework's core functionality, and its accuracy forms the basis for subsequent validation.

[0129] Based on this stratification result, a high-fidelity digital simulation scenario is constructed in a computing environment. This scenario maps the stratification result into a three-dimensional structure of a virtual soil profile, where different soil layers are assigned corresponding physicochemical parameters extracted from the stratified soil property dataset, such as porosity, density, and pH value, thereby accurately reproducing the vertical heterogeneity of the target soil. Subsequently, in the simulation scenario, a virtual tracer is injected in a software simulation manner, and an embedded infiltration path prediction model (such as a random forest model) derived from step 4 is invoked to dynamically reproduce and track the entire lifecycle of the amendment from injection and infiltration to action. This reproduction and tracking process is based on physical laws and machine learning models, simulating and calculating the diffusion rate of the virtual tracer in different soil layers, predicting the possible hindrance points it may encounter and their evolution.

[0130] Based on the simulation data generated during the re-tracking process, the system executes a parallel processing flow to step 5, extracting the simulated duration and intensity of the soil amendment's action on each soil layer. These data are then used as the effect data for each soil layer. The duration of action is extracted based on the variation curve of the virtual tracer concentration over the simulated time series, calculating the duration exceeding a specific threshold. The calculation of the effect intensity combines the simulated diffusion rate distribution with the duration of action, and similarly incorporates a weighting coefficient based on soil layer depth.

[0131] Finally, a comprehensive analysis is performed on the extracted data representing the predicted effects of the framework. This analysis includes calculating the average value of the simulated effects for each soil layer, determining the gradient differences in effects between layers, and comparing the simulated effects with preset improvement targets or historical benchmarks. Based on the conclusions of this comprehensive analysis, a vertical stratified effect assessment report on the soil improvement is automatically generated. This report, in document and chart form, quantitatively demonstrates the expected effects of the amendment in the vertical direction and in each soil layer, clearly identifying the areas of advantage and potential weaknesses in the improvement.

[0132] This data processing and verification process addresses the technical challenges of ambiguous assessments of soil improvement effects and insufficient decision-making basis caused by a lack of reliable evaluation tools. By systematically testing an integrated and intelligent monitoring framework using independent real-world data, and reproducing and predicting the entire improvement process in a simulated environment, it ultimately generates a quantitative assessment report based on data-driven and model-predictive methods. This process elevates the assessment of soil improvement effects from qualitative judgments relying on experience to quantitative predictions based on model simulation and data analysis. It provides managers with clear and forward-looking insights into the potential vertically stratified effects of improvement measures, thereby supporting optimized decision-making and enhancing the scientific rigor and success rate of soil improvement projects.

[0133] The above describes the method for real-time monitoring of soil improvement effects in the embodiments of this application. The following describes the system for real-time monitoring of soil improvement effects in the embodiments of this application. Please refer to [link / reference]. Figure 2 The present application provides a schematic diagram of the structure of a real-time monitoring system for soil improvement effects, which includes:

[0134] The data acquisition module is used to collect multi-depth data samples of soil in the vertical direction through a sensor array to construct a layered soil characteristic dataset.

[0135] The segment identification module is used to identify the location of highly heterogeneous segments in the soil based on the physical and chemical properties of the stratified soil characteristic dataset.

[0136] The image acquisition module is used to inject tracers into highly heterogeneous sections to simulate the flow of modifiers and capture their distribution images in real time to obtain a sequence of permeation path images.

[0137] The path optimization module is used to predict the diffusion rate and hindrance point location of the tracer based on the permeation path image sequence, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier and obtain updated permeation path data.

[0138] The effect quantification module is used to track and quantify the changes in the effect of the amendment in each soil layer based on the updated infiltration path data, and obtain quantitative indicators of the effect.

[0139] The model update module is used to compare the quantitative indicators of the effect with the stratified soil characteristic dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring.

[0140] The validation report module is used to integrate multi-depth data samples to validate the hierarchical monitoring framework and generate a final vertical hierarchical effect evaluation report.

[0141] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the real-time monitoring method for soil improvement effects.

[0142] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for real-time monitoring of soil amelioration effects, characterized in that, The method includes: Step 1: Collect multi-depth data samples of soil in the vertical direction using a sensor array to construct a layered soil characteristic dataset; Step 2: Based on the physical and chemical properties of the stratified soil characteristic dataset, identify the locations of highly heterogeneous zones in the soil; Step 3: Inject tracers into the high heterogeneous section to simulate the flow of modifiers and capture their distribution images in real time to obtain a permeation path image sequence; Step 4: Based on the permeation path image sequence, predict the diffusion rate and stagnation point location of the tracer, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier, and obtain updated permeation path data. Step 5: Based on the updated infiltration path data, track and quantify the changes in the effect of the amendment in each soil layer, and obtain quantitative indicators of the effect. Step 6: Compare the quantitative indicators of the effect with the stratified soil characteristic dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring accordingly. Step 7: Integrate the multi-depth data samples to verify the hierarchical monitoring framework and generate the final vertical hierarchical effect evaluation report; The process includes comparing the quantitative indicators of the effects with the stratified soil characteristic dataset, and optimizing the soil stratification model based on the comparison results, including: All initial characteristic values ​​of any soil layer are read to form a feature vector. These initial characteristic values ​​are input into a prediction function to map multiple initial characteristic values ​​to a single predicted effect value. The predicted effect value is compared with the quantitative index of the effect of the soil layer to obtain a relative deviation value. The relative deviation values ​​corresponding to each layer are weighted and summed to obtain the difference value between the quantitative index of the effect and the initial characteristic value. If the difference value is higher than a preset optimization threshold, the soil stratification model is iteratively optimized to obtain an optimized stratification model.

2. The method of claim 1, wherein, Step 1 includes: A stratified sampling method was used to divide the vertical soil profile into topsoil, subsoil and subsoil, and data samples of each soil layer were collected by a sensor array. The data samples were preprocessed, and then the physical and chemical parameters of each soil layer were extracted. The physical parameters and the chemical parameters are used to construct an initial layered soil property dataset, which contains the property distribution information of each soil layer in the vertical direction. The initial stratified soil property dataset is standardized to obtain structured standard data; The structured standard data is stored as a multidimensional matrix, where rows of the matrix represent different soil layers and columns represent different parameters; Feature extraction is performed on the multidimensional matrix to determine the correlation between the properties of each soil layer; The initial stratified soil property dataset is preliminarily stratified and validated based on the correlation. If the validation is successful, the structured standard data is used as the stratified soil property dataset.

3. The method of claim 1, wherein, Step 2 includes: The layered soil property dataset is input into a pre-trained support vector machine model. The support vector machine model is used to classify the physical properties and chemical composition in the dataset to obtain the classification results of each soil layer. Based on the classification results, calculate the characteristic differences in physical and chemical properties between adjacent soil layers; Determine whether the characteristic difference value exceeds a preset difference threshold. If so, mark the corresponding soil segment as a highly heterogeneous segment. For the marked highly heterogeneous segments, extract their vertical position information; A distribution map of highly heterogeneous segments is generated based on the location information. The highly heterogeneous segments are then accurately located using the distribution map to obtain their positions.

4. The method of claim 1, wherein, Step 3 includes: The tracer injection point is determined based on the location of the highly heterogeneous segment; The tracer is injected at the injection point using an injection device to simulate the flow of the amendment in the vertical profile of the soil. The distribution of the tracer in the vertical path is dynamically captured using a real-time imaging device, and multiple distribution images are obtained based on the dynamically captured data. The distribution images are arranged in time sequence to construct an initial permeation path image sequence; The initial permeation path image sequence is denoised to obtain the permeation path image sequence.

5. The method of claim 1, wherein, Step 4 includes: The permeation path image sequence is input into a pre-trained random forest model, and the random forest model extracts features from the permeation path image sequence to predict the diffusion rate of the tracer. The flow resistance of the tracer in each soil layer is determined based on the diffusion rate. Determine whether the flow resistance exceeds a preset resistance threshold; if so, determine that there is a blockage point. For the identified obstruction points, their coordinates are extracted from the permeation path image sequence to obtain the location of the obstruction points; If the blockage point location information shows that the blockage point is located in the bottom soil layer, then adjust the injection parameters, which include injection pressure and injection rate; The tracer was re-injected according to the adjusted injection parameters, and a new permeation path image sequence was captured. The new permeation path image sequence is analyzed to determine whether the blockage point has been eliminated. If so, the new permeation path image sequence is used as the updated permeation path data. If not, the injection parameters are iteratively adjusted until the blockage point is eliminated.

6. The method of claim 1, wherein, Step 5 includes: The updated permeation path data was processed using time-series analysis to extract the distribution characteristics of the amendment in each soil layer. Based on the distribution characteristics, the duration of action of the soil conditioner in each soil layer is calculated; By performing a weighted analysis on the duration of action, the intensity of action of the amendment in each soil layer was determined; For each soil layer, obtain the corresponding quantitative value of the effect based on its corresponding action intensity and duration; The quantitative values ​​of the effects corresponding to each soil layer are standardized to obtain the final quantitative index of the effects of each soil layer.

7. The method according to claim 1, characterized in that, Step 6 includes: The quantitative indicators of the effects of each soil layer are aligned layer by layer with the initial characteristic values ​​of the corresponding soil layer in the stratified soil characteristic dataset. Calculate the difference between the quantitative indicator of the effect and the initial characteristic value; If the difference value is higher than the preset optimization threshold, the soil stratification model is iteratively optimized to obtain the optimized stratification model. A layered monitoring framework is constructed based on the optimized layered model. The layered monitoring framework includes monitoring modules corresponding to each soil layer and a real-time data input interface. The applicability of the hierarchical monitoring framework was determined by functional verification. The validated stratified monitoring framework is associated and stored with the stratified soil characteristic dataset, and a data synchronization and update mechanism is established. A hierarchical monitoring strategy is generated by mapping the data of the hierarchical monitoring framework.

8. The method according to claim 1, characterized in that, Step 7 includes: The multi-depth data samples are input into a validated hierarchical monitoring framework; The multi-depth data samples are analyzed by the layered monitoring framework to generate layered results that include the distribution of soil layer characteristics and the location of highly heterogeneous sections. Based on the hierarchical results, a simulation scenario is constructed, and a virtual tracer is injected into the simulation scenario to reproduce and track the entire process of the modifier's action. Based on the reproduction and tracking process, the simulated duration and intensity of the amendment in each soil layer were extracted as data on the effect of the amendment in each soil layer. A comprehensive analysis of the aforementioned effect data is conducted to generate a vertical stratification effect assessment report on the soil improvement effect.

9. A real-time monitoring system for soil improvement effects, used to implement the method as described in any one of claims 1 to 8, characterized in that, The system includes: The data acquisition module is used to collect multi-depth data samples of soil in the vertical direction through a sensor array to construct a layered soil characteristic dataset; The segment identification module is used to identify the location of highly heterogeneous segments in the soil based on the physical and chemical properties of the layered soil characteristic dataset. The image acquisition module is used to inject tracers at the location of the highly heterogeneous section to simulate the flow of modifiers and capture their distribution images in real time to obtain a permeation path image sequence. The path optimization module is used to predict the diffusion rate and hindrance point location of the tracer based on the permeation path image sequence, and dynamically adjust the injection parameters based on the prediction results to optimize the permeation effect of the modifier and obtain updated permeation path data. The effect quantification module is used to track and quantify the changes in the effect of the amendment in each soil layer based on the updated infiltration path data, and obtain quantitative indicators of the effect. The model update module is used to compare the quantitative indicators of the effect with the stratified soil characteristic dataset, optimize the soil stratification model based on the comparison results, and determine the stratified monitoring framework for real-time monitoring accordingly. The verification report module is used to integrate the multi-depth data samples to verify the hierarchical monitoring framework and generate a final vertical hierarchical effect evaluation report.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method for real-time monitoring of soil improvement effects as described in any one of claims 1 to 8.

Images